A CFD-based surrogate model for predicting flow parameters in a ventilated room using sensor readings

Abstract

In this work, we develop a computational fluid dynamics (CFD)-based surrogate model, which predicts flow parameters under different geometrical configurations and boundary conditions in a benchmark case of a mechanically ventilated room with mixed convection. The model inputs are the temperature and velocity values in different locations, which act as a surrogate of the sensor readings. The model’s output is a set of comfort-related flow parameters, such as the average Nusselt number on the hot wall, jet separation point, average kinetic energy, average enstrophy, and average temperature. We tested four different machine learning methods, among which we chose the gradient boosting regression due to its accurate performance. We also adapted the developed model for indoor environment control applications by determining the optimal combinations of sensor positions which minimize the prediction error. This model does not require the repetition of CFD simulations in order to be applied since the structure of the input data imitates sensor readings. Furthermore, the low computational cost of the model execution and good accuracy makes it an effective alternative to CFD for applications where rapid predictions of complex flow configurations are required, such as model predictive control.